Identifying Necessary Components for Open Ended Evolution

Anya Vostinar, Emily Dolson, Michael Wiser, and Charles Ofria

June 3rd, 2016

Open-Ended Evolution is a huge concept

  • To make scientific progress, we need a way to approach it incrementally
  • Metrics of relative open-endedness allow this
  • Complexity barriers

Our approach

  • Last year we presented a suite of four metrics:
    • Change
    • Novelty
    • Ecology
    • Complexity
  • Here, we test these metrics in a simple, well-studied system: NK landscapes

NK Landscapes

  • Popular model for studying evolutionary dynamics in bitstrings
  • N = length of bitstring
  • K = Epistasis

1 0 1 1 1 0 0

Filtering out noise

Filtering out noise

  • Evolution is an inherently noisy process
    • Not all parts of a genome contribute to its success
    • Many members of a population are the result of deleterious mutations

Filtering the genome

  • Determine the fitness effect of changing each site in genome
  • If negative, that site is informative - keep it
  • Ignore all other sites

Example goes here

Filtering the population

  • Previous approaches:
    • Evolutionary activity statistics shadow run
    • Fossil record
  • We build on Bedau et. al.’s approach to the fossil record

Metrics

Change

  • Do the strategies in the population keep changing?
  • In may EC systems, they don’t

Novelty

  • Do new strategies keep appearing?

Ecology

  • Does the diversity of strategies in the population keep increasing?

Complexity

  • Does the maximum complexity in the population keep increasing?

A simple example

K example

Pop size example

Changing environments example

Fitness sharing example

Results

Change

Change

Novelty

Novelty

Ecology

Ecology

Complexity

Complexity

Conclusions

Conclusions

  • Metrics intuitively reflect evolutionary dynamics
  • By measuring the effects of different treaments, we can zero in on which conditions are necessary for open-ended evolution

Acknowledgements

Co-authors:

Funding sources: